Verified: a year ago
SafeTensor
Type | |
Stats | 3,166 47,672 |
Reviews | (317) |
Published | Oct 13, 2023 |
Base Model | |
Training | Steps: 2,000 Epochs: 20 |
Hash | AutoV2 0C73F1D83B |
2023.10.13 Update - SDXL Version
It works well with BluePencil, Counterfeit, Reproduction and Breakdomain. You can try any anime or realistic XL models!
Still recommended to use lora weight 1.
CFG between 3.5~7, value too high will cause color oversaturated.
Recommended prompts if you want to more close to original style:
hair ornament, cube hair ornament, blue eyes, pink long hair, pink track jacket, bangs, hair between eyes
Enjoy your bocchi chan!
2023.2.9 First version base on SD 1.5
To reduce overfitting, I cut down training epochs, removed trigger words (it's not necessary) and set Network Rank And Network Alpha to 64.
Have good results when use LoRA weights 1.
Sample images generated by NovelAI, Anything 4.5 and AbyssOrange.